Input data and preprocessing

Input Data & Preprocessing

📥 Input Format

adabmDCA 2.0 takes as input a multiple sequence alignment (MSA) in FASTA format, typically of aligned protein or RNA/DNA sequences (see Fig. 1).

The tool supports three built-in alphabets and also allows custom alphabets, as long as they match the MSA content.

Type	Alphabet Symbols
protein	-, A, C, D, E, F, G, H, I, K, L, M, N, P, Q, R, S, T, V, W, Y
RNA	-, A, C, G, U
DNA	-, A, C, G, T

💡 Line breaks within sequences are supported.

Figure 1: Example of a fasta file containg the MSA.

---

🔧 Preprocessing Steps

The following steps are applied to every input MSA:

1. Remove sequences with invalid symbols.
2. Remove duplicate sequences.
3. Reweight sequences to correct for phylogenetic and sampling bias (optional).
4. Compute empirical statistics using a pseudocount.

---

⚖️ Sequence Reweighting

To downweight overrepresented or phylogenetically related sequences, adabmDCA uses a clustering threshold (default: 80% identity). The weight for sequence \(\mathbf{a}^{(m)}\) is:

\[ w^{(m)} = \frac{1}{N^{(m)}} \]

where \(N^{(m)}\) is the number of sequences that have sequence identity with \(\mathbf{a}^{(m)}\) above the clustering threshold.

• Set the threshold with --clustering_seqid <value>
• Disable with --no_reweighting

---

🧮 Pseudocount Regularization

A small pseudocount \(\alpha\) is added to frequency estimates to prevent issues with rare or unobserved symbols:

• One-site frequency:
  \(f_i(a) = (1 - \alpha) f^{\mathrm{data}}_i(a) + \frac{\alpha}{q}\)

• Two-site frequency:
  \(f_{ij}(a, b) = (1 - \alpha) f^{\mathrm{data}}_{ij}(a, b) + \frac{\alpha}{q^2}\)

If not set via --pseudocount, the default is: $$ \alpha = \frac{1}{M_{\text{eff}}}, \quad \text{with} \quad M_{\text{eff}} = \sum_{m=1}^M w^{(m)} $$ being the effective number of sequences.

---